A Brief Introduction to Operating Systems

Citation preview

THINK OS - A BRIEF INTRODUCTION TO OPERATING SYSTEMS

Allen B. Downey Olin College

Default Text

This text is disseminated via the Open Education Resource (OER) LibreTexts Project (https://LibreTexts.org) and like the hundreds of other texts available within this powerful platform, it is freely available for reading, printing and "consuming." Most, but not all, pages in the library have licenses that may allow individuals to make changes, save, and print this book. Carefully consult the applicable license(s) before pursuing such effects. Instructors can adopt existing LibreTexts texts or Remix them to quickly build course-specific resources to meet the needs of their students. Unlike traditional textbooks, LibreTexts’ web based origins allow powerful integration of advanced features and new technologies to support learning.

The LibreTexts mission is to unite students, faculty and scholars in a cooperative effort to develop an easy-to-use online platform for the construction, customization, and dissemination of OER content to reduce the burdens of unreasonable textbook costs to our students and society. The LibreTexts project is a multi-institutional collaborative venture to develop the next generation of openaccess texts to improve postsecondary education at all levels of higher learning by developing an Open Access Resource environment. The project currently consists of 14 independently operating and interconnected libraries that are constantly being optimized by students, faculty, and outside experts to supplant conventional paper-based books. These free textbook alternatives are organized within a central environment that is both vertically (from advance to basic level) and horizontally (across different fields) integrated. The LibreTexts libraries are Powered by NICE CXOne and are supported by the Department of Education Open Textbook Pilot Project, the UC Davis Office of the Provost, the UC Davis Library, the California State University Affordable Learning Solutions Program, and Merlot. This material is based upon work supported by the National Science Foundation under Grant No. 1246120, 1525057, and 1413739. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation nor the US Department of Education. Have questions or comments? For information about adoptions or adaptions contact [email protected]. More information on our activities can be found via Facebook (https://facebook.com/Libretexts), Twitter (https://twitter.com/libretexts), or our blog (http://Blog.Libretexts.org). This text was compiled on 08/24/2023

TABLE OF CONTENTS Licensing

0: Preface 0.1: Preface 0.2: Using the code 0.3: Contributor List

1: Compilation 1.1: Compiled and interpreted languages 1.2: Static types 1.3: The compilation process 1.4: Object code 1.5: Assembly code 1.6: Preprocessing 1.7: Understanding errors

2: Processes 2.1: Abstraction and virtualization 2.2: Isolation 2.3: UNIX processes

3: Virtual memory 3.1: A bit of information theory 3.2: Memory and storage 3.3: Address spaces 3.6: Address translation 3.4: Memory segments 3.5: Static local variables

4: Files and file systems 4.1: Files and file systems 4.2: Disk performance 4.3: Disk metadata 4.4: Block allocation 4.5: Everything is a file?

5: More bits and bytes 5.1: Representing integers 5.2: Bitwise operators 5.3: Representing floating-point numbers 5.4: Unions and memory errors 5.5: Representing strings

1

https://eng.libretexts.org/@go/page/43661

6: Memory management 6.1: Memory errors 6.2: Memory leaks 6.3: Implementation

7: Caching 7.1: How programs run 7.2: Cache performance 7.3: Locality 7.4: Measuring cache performance 7.5: Programming for cache performance 7.6: The memory hierarchy 7.7: Caching policy 7.8: Paging

8: Multitasking 8.1: Hardware state 8.2: Context switching 8.3: The process life cycle 8.4: Scheduling 8.5: Real-time scheduling

9: Threads 9.1: Creating threads 9.2: Creating threads 9.3: Joining threads 9.4: Synchronization errors 9.5: Mutex

10: Condition variables 10.1: The work queue 10.2: Producers and consumers 10.3: Mutual exclusion 10.4: Condition variables 10.5: Condition variable implementation

11: Semaphores in C 11.1: POSIX Semaphores 11.2: Producers and consumers with semaphores 11.3: Make your own semaphores

Index Glossary Detailed Licensing

2

https://eng.libretexts.org/@go/page/43661

Licensing A detailed breakdown of this resource's licensing can be found in Back Matter/Detailed Licensing.

1

https://eng.libretexts.org/@go/page/96722

CHAPTER OVERVIEW 0: Preface 0.1: Preface 0.2: Using the code 0.3: Contributor List

This page titled 0: Preface is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1

0.1: Preface In many computer science programs, Operating Systems is an advanced topic. By the time students take it, they know how to program in C, and they have probably taken a class in Computer Architecture. Usually the goal of the class is to expose students to the design and implementation of operating systems, with the implied assumption that some of them will do research in this area, or write part of an OS. This book is intended for a different audience, and it has different goals. I developed it for a class at Olin College called Software Systems. Most students taking this class learned to program in Python, so one of the goals is to help them learn C. For that part of the class, I use Griffiths and Griffiths, Head First C, from O’Reilly Media. This book is meant to complement that one. Few of my students will ever write an operating system, but many of them will write low-level applications in C or work on embedded systems. My class includes material from operating systems, networks, databases, and embedded systems, but it emphasizes the topics programmers need to know. This book does not assume that you have studied Computer Architecture. As we go along, I will explain what we need. If this book is successful, it should give you a better understanding of what is happening when programs run, and what you can do to make them run better and faster. Chapter 1 explains some of the differences between compiled and interpreted languages, with some insight into how compilers work. Recommended reading: Head First C Chapter 1. Chapter 2 explains how the operating system uses processes to protect running programs from interfering with each other. Chapter 3 explains virtual memory and address translation. Recommended reading: Head First C Chapter 2. Chapter 4 is about file systems and data streams. Recommended reading: Head First C Chapter 3. Chapter 5 describes how numbers, letters, and other values are encoded, and presents the bitwise operators. Chapter 6 explains how to use dynamic memory management, and how it works. Recommended reading: Head First C Chapter 6. Chapter 7 is about caching and the memory hierarchy. Chapter 8 is about multitasking and scheduling. Chapter 9 is about POSIX threads and mutexes. Recommended reading: Head First C Chapter 12 and Little Book of Semaphores Chapters 1 and 2. Chapter 10 is about POSIX condition variables and the producer/consumer problem. Recommended reading: Little Book of Semaphores Chapters 3 and 4. Chapter 11 is about using POSIX semaphores and implementing semaphores in C.

A note on this draft The current version of this book is an early draft. While I am working on the text, I have not yet included the figures. So there are a few places where, I’m sure, the explanation will be greatly improved when the figures are ready. This page titled 0.1: Preface is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

0.1.1

https://eng.libretexts.org/@go/page/42832

0.2: Using the code Example code for this book is available from https://github.com/AllenDowney/ThinkOS. Git is a version control system that allows you to keep track of the files that make up a project. A collection of files under Git’s control is called a repository. GitHub is a hosting service that provides storage for Git repositories and a convenient web interface. The GitHub homepage for my repository provides several ways to work with the code: You can create a copy of my repository on GitHub by pressing the Fork button. If you don’t already have a GitHub account, you’ll need to create one. After forking, you’ll have your own repository on GitHub that you can use to keep track of code you write while working on this book. Then you can clone the repo, which means that you copy the files to your computer. Or you could clone my repository. You don’t need a GitHub account to do this, but you won’t be able to write your changes back to GitHub. If you don’t want to use Git at all, you can download the files in a Zip file using the button in the lower-right corner of the GitHub page. This page titled 0.2: Using the code is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

0.2.1

https://eng.libretexts.org/@go/page/42835

0.3: Contributor List If you have a suggestion or correction, please send email to [email protected]. If I make a change based on your feedback, I will add you to the contributor list (unless you ask to be omitted). If you include at least part of the sentence the error appears in, that makes it easy for me to search. Page and section numbers are fine, too, but not quite as easy to work with. Thanks! I am grateful to the students in Software Systems at Olin College, who tested an early draft of this book in Spring 2014. They corrected many errors and made many helpful suggestions. I appreciate their pioneering spirit! James P Giannoules spotted a copy-and-paste error. Andy Engle knows the difference between GB and GiB. Aashish Karki noted some broken syntax. Other people who found typos and errors include Jim Tyson, Donald Robertson, Jeremy Vermast, Yuzhong Huang, Ian Hill. This page titled 0.3: Contributor List is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

0.3.1

https://eng.libretexts.org/@go/page/42836

CHAPTER OVERVIEW 1: Compilation 1.1: Compiled and interpreted languages 1.2: Static types 1.3: The compilation process 1.4: Object code 1.5: Assembly code 1.6: Preprocessing 1.7: Understanding errors

This page titled 1: Compilation is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1

1.1: Compiled and interpreted languages People often describe programming languages as either compiled or interpreted. “Compiled” means that programs are translated into machine language and then executed by hardware; “interpreted” means that programs are read and executed by a software interpreter. Usually C is considered a compiled language and Python is considered an interpreted language. But the distinction is not always clear-cut. First, many languages can be either compiled or interpreted. For example, there are C interpreters and Python compilers. Second, there are languages like Java that use a hybrid approach, compiling programs into an intermediate language and then running the translated program in an interpreter. Java uses an intermediate language called Java bytecode, which is similar to machine language, but it is executed by a software interpreter, the Java virtual machine (JVM). So being compiled or interpreted is not an intrinsic characteristic of a language; nevertheless, there are some general differences between compiled and interpreted languages. This page titled 1.1: Compiled and interpreted languages is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1.1.1

https://eng.libretexts.org/@go/page/40627

1.2: Static types Many interpreted languages support dynamic types, but compiled languages are usually limited to static types. In a statically-typed language, you can tell by looking at the program what type each variable refers to. In a dynamically-typed language, you don’t always know the type of a variable until the program is running. In general, static refers to things that happen at compile time (while a program is being compiled), and dynamic refers to things that happen at run time (while a program is running). For example, in Python you can write a function like this: def add(x, y): return x + y Looking at this code, you can’t tell what type x and y will refer to at run time. This function might be called several times, each time with values with different types. Any values that support the addition operator will work; any other types will cause an exception or runtime error. In C you would write the same function like this: int add(int x, int y) { return x + y; } The first line of the function includes type declarations for the parameters and the return value: x and y are declared to be integers, which means that we can check at compile time whether the addition operator is legal for this type (it is). The return value is also declared to be an integer. Because of these declarations, when this function is called elsewhere in the program, the compiler can check whether the arguments provided have the right type, and whether the return value is used correctly. These checks happen before the program starts executing, so errors can be found earlier. More importantly, errors can be found in parts of the program that have never run. Furthermore, these checks don’t have to happen at run time, which is one of the reasons compiled languages generally run faster than interpreted languages. Declaring types at compile time also saves space. In dynamic languages, variable names are stored in memory while the program runs, and they are often accessible by the program. For example, in Python the built-in function locals returns a dictionary that contains variable names and their values. Here’s an example in a Python interpreter: >>> x = 5 >>> print locals() {'x': 5, '__builtins__': , '__name__': '__main__', '__doc__': None, '__package__': None} This shows that the name of the variable is stored in memory while the program is running (along with some other values that are part of the default runtime environment). In compiled languages, variable names exist at compile-time but not at run time. The compiler chooses a location for each variable and records these locations as part of the compiled program.1 The location of a variable is called its address. At run time, the value of each variable is stored at its address, but the names of the variables are not stored at all (unless they are added by the compiler for purposes of debugging). 1. This is a simplification; we will go into more detail later. This page titled 1.2: Static types is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1.2.1

https://eng.libretexts.org/@go/page/40628

1.3: The compilation process As a programmer, you should have a mental model of what happens during compilation. If you understand the process, it will help you interpret error messages, debug your code, and avoid common pitfalls. The steps of compilation are: 1. Preprocessing: C is one of several languages that include preprocessing directives that take effect before the program is compiled. For example, the #include directive causes the source code from another file to be inserted at the location of the directive. 2. Parsing: During parsing, the compiler reads the source code and builds an internal representation of the program, called an abstract syntax tree. Errors detected during this step are generally syntax errors. 3. Static checking: The compiler checks whether variables and values have the right type, whether functions are called with the right number and type of arguments, etc. Errors detected during this step are sometimes called static semantic errors. 4. Code generation: The compiler reads the internal representation of the program and generates machine code or byte code. 5. Linking: If the program uses values and functions defined in a library, the compiler has to find the appropriate library and include the required code. 6. Optimization: At several points in the process, the compiler can transform the program to generate code that runs faster or uses less space. Most optimizations are simple changes that eliminate obvious waste, but some compilers perform sophisticated analyses and transformations. Normally when you run gcc , it runs all of these steps and generates an executable file. For example, here is a minimal C program: #include int main() { printf("Hello World\n"); } If you save this code in a file called hello.c , you can compile and run it like this: $ gcc hello.c $ ./a.out By default, gcc stores the executable code in a file called a.out (which originally stood for “assembler output”). The second line runs the executable. The prefix ./ tells the shell to look for it in the current directory. It is usually a good idea to use the -o flag to provide a better name for the executable: $ gcc hello.c -o hello $ ./hello This page titled 1.3: The compilation process is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1.3.1

https://eng.libretexts.org/@go/page/40629

1.4: Object code The -c flag tells gcc to compile the program and generate machine code, but not to link it or generate an executable: $ gcc hello.c -c The result is a file named hello.o , where the o stands for object code, which is the compiled program. Object code is not executable, but it can be linked into an executable. The UNIX command nm reads an object file and generates information about the names it defines and uses. For example: $ nm hello.o 0000000000000000 T main U puts This output indicates that hello.o defines the name main and uses a function named puts , which stands for “put string”. In this example, gcc performs an optimization by replacing printf , which is a large and complicated function, with puts , which is relatively simple. You can control how much optimization gcc does with the -O flag. By default, it does very little optimization, which can make debugging easier. The option -O1 turns on the most common and safe optimizations. Higher numbers turn on additional optimizations that require longer compilation time. In theory, optimization should not change the behavior of the program, other than to speed it up. But if your program has a subtle bug, you might find that optimization makes the bug appear or disappear. It is usually a good idea to turn off optimization while you are developing new code. Once the program is working and passing appropriate tests, you can turn on optimization and confirm that the tests still pass. This page titled 1.4: Object code is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1.4.1

https://eng.libretexts.org/@go/page/40630

1.5: Assembly code Similar to the -c flag, the -S flag tells gcc to compile the program and generate assembly code, which is basically a human-readable form of machine code. $ gcc hello.c -S The result is a file named hello.s , which might look something like this: .file .section

"hello.c" .rodata

.string .text .globl .type

"Hello World"

.LC0:

main main, @function

main: .LFB0: .cfi_startproc pushq %rbp .cfi_def_cfa_offset 16 .cfi_offset 6, -16 movq %rsp, %rbp .cfi_def_cfa_register 6 movl $.LC0, %edi call puts movl $0, %eax popq %rbp .cfi_def_cfa 7, 8 ret .cfi_endproc .LFE0: .size .ident .section

main, .-main "GCC: (Ubuntu/Linaro 4.7.3-1ubuntu1) 4.7.3" .note.GNU-stack,"",@progbits

gcc is usually configured to generate code for the machine you are running on, so for me it generates x86 assembly language, which runs on a wide variety of processors from Intel, AMD, and others. If you are running on a different architecture, you might see different code. This page titled 1.5: Assembly code is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1.5.1

https://eng.libretexts.org/@go/page/40631

1.6: Preprocessing Taking another step backward through the compilation process, you can use the -E flag to run the preprocessor only: $ gcc hello.c -E The result is the output from the preprocessor. In this example, it contains the included code from stdio.h , and all the files included from stdio.h , and all the files included from those files, and so on. On my machine, the total is more than 800 lines of code. Since almost every C program includes stdio.h , those 800 lines of code get compiled a lot. If, like many C programs, you also include stdlib.h , the result is more than 1800 lines of code. This page titled 1.6: Preprocessing is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1.6.1

https://eng.libretexts.org/@go/page/40632

1.7: Understanding errors Now that we know the steps in the compilation process, it is easier to understand error messages. For example, if there is an error in a #include directive, you’ll get a message from the preprocessor: hello.c:1:20: fatal error: stdioo.h: No such file or directory compilation terminated. If there’s a syntax error, you get a message from the compiler: hello.c: In function 'main': hello.c:6:1: error: expected ';' before '}' token If you use a function that’s not defined in any of the standard libraries, you get a message from the linker: /tmp/cc7iAUbN.o: In function `main': hello.c:(.text+0xf): undefined reference to `printff' collect2: error: ld returned 1 exit status ld is the name of the UNIX linker, so named because “loading” is another step in the compilation process that is closely related to linking. Once the program starts, C does very little runtime checking, so there are only a few runtime errors you are likely to see. If you divide by zero, or perform another illegal floating-point operation, you will get a “Floating point exception.” And if you try to read or write an incorrect location in memory, you will get a “Segmentation fault.” This page titled 1.7: Understanding errors is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1.7.1

https://eng.libretexts.org/@go/page/42844

CHAPTER OVERVIEW 2: Processes 2.1: Abstraction and virtualization 2.2: Isolation 2.3: UNIX processes

This page titled 2: Processes is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1

2.1: Abstraction and virtualization Before we talk about processes, I want to define a few words: Abstraction: An abstraction is a simplified representation of something complicated. For example, if you drive a car, you understand that when you turn the wheel left, the car goes left, and vice versa. Of course, the steering wheel is connected to a sequence of mechanical and (often) hydraulic systems that turn the wheels, and the wheels interact with the road in ways that can be complex, but as a driver, you normally don’t have to think about any of those details. You can get along very well with a simple mental model of steering. Your mental model is an abstraction. Similarly, when you use a web browser, you understand that when you click on a link, the browser displays the page the link refers to. The software and network communication that make that possible are complex, but as a user, you don’t have to know the details. A large part of software engineering is designing abstractions like these that allow users and other programmers to use powerful and complicated systems without having to know about the details of their implementation. Virtualization: An important kind of abstraction is virtualization, which is the process of creating a desirable illusion. For example, many public libraries participate in inter-library collaborations that allow them to borrow books from each other. When I request a book, sometimes the book is on the shelf at my local library, but other times it has to be transferred from another collection. Either way, I get a notification when it is available for pickup. I don’t need to know where it came from, and I don’t need to know which books my library has. As a whole, the system creates the illusion that my library has every book in the world. The collection physically located at my local library might be small, but the collection available to me virtually includes every book in the inter-library collaboration. As another example, most computers are only connected to one network, but that network is connected to others, and so on. What we call the Internet is a collection of networks and a set of protocols that forward packets from one network to the next. From the point of view of a user or programmer, the system behaves as if every computer on the Internet is connected to every other computer. The number of physical connections is small, but the number of virtual connections is very large. The word “virtual” is often used in the context of a virtual machine, which is software that creates the illusion of a dedicated computer running a particular operating system, when in reality the virtual machine might be running, along with many other virtual machines, on a computer running a different operating system. In the context of virtualization, we sometimes call what is really happening “physical”, and what is virtually happening either “logical” or “abstract.” This page titled 2.1: Abstraction and virtualization is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

2.1.1

https://eng.libretexts.org/@go/page/40704

2.2: Isolation One of the most important principles of engineering is isolation: when you are designing a system with multiple components, it is usually a good idea to isolate them from each other so that a change in one component doesn’t have undesired effects on other components. One of the most important goals of an operating system is to isolate each running program from the others so that programmers don’t have to think about every possible interaction. The software object that provides this isolation is a process. A process is a software object that represents a running program. I mean “software object” in the sense of object-oriented programming; in general, an object contains data and provides methods that operate on the data. A process is an object that contains the following data: The text of the program, usually a sequence of machine language instructions. Data associated with the program, including static data (allocated at compile time) and dynamic data (allocated at run time). The state of any pending input/output operations. For example, if the process is waiting for data to be read from disk or for a packet to arrive on a network, the status of these operations is part of the process. The hardware state of the program, which includes data stored in registers, status information, and the program counter, which indicates which instruction is currently executing. Usually one process runs one program, but it is also possible for a process to load and run a new program. It is also possible, and common, to run the same program in more than one process. In that case, the processes share the same program text but generally have different data and hardware states. Most operating systems provide a fundamental set of capabilities to isolate processes from each other: Multitasking: Most operating systems have the ability to interrupt a running process at almost any time, save its hardware state, and then resume the process later. In general, programmers don’t have to think about these interruptions. The program behaves as if it is running continuously on a dedicated processor, except that the time between instructions is unpredictable. Virtual memory: Most operating systems create the illusion that each process has its own chunk of memory, isolated from all other processes. Again, programmers generally don’t have to think about how virtual memory works; they can proceed as if every program has a dedicated chunk of memory. Device abstraction: Processes running on the same computer share the disk drive, the network interface, the graphics card, and other hardware. If processes interacted with this hardware directly, without coordination, chaos would ensue. For example, network data intended for one process might be read by another. Or multiple processes might try to store data in the same location on a hard drive. It is up to the operating system to maintain order by providing appropriate abstractions. As a programmer, you don’t need to know much about how these capabilities are implemented. But if you are curious, you will find a lot of interesting things going on under the metaphorical hood. And if you know what’s going on, it can make you a better programmer. This page titled 2.2: Isolation is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

2.2.1

https://eng.libretexts.org/@go/page/40705

2.3: UNIX processes While I write this book, the process I am most aware of is my text editor, emacs. Every once in a while I switch to a terminal window, which is a window running a UNIX shell that provides a command-line interface. When I move the mouse, the window manager wakes up, sees that the mouse is over the terminal window, and wakes up the terminal. The terminal wakes up the shell. If I type make in the shell, it creates a new process to run Make, which creates another process to run LaTeX and then another process to display the results. If I need to look something up, I might switch to another desktop, which wakes up the window manager again. If I click on the icon for a web browser, the window manager creates a process to run the web browser. Some browsers, like Chrome, create a new process for each window and each tab. And those are just the processes I am aware of. At the same time there are many other processes running in the background. Many of them are performing operations related to the operating system. The UNIX command ps prints information about running processes. If you run it in a terminal, you might see something like this: PID 2687 2801 24762

TTY pts/1 pts/1 pts/1

TIME 00:00:00 00:01:24 00:00:00

CMD bash emacs ps

The first column is the unique numerical process ID. The second column is the terminal that created the process; “TTY” stands for teletypewriter, which was the original mechanical terminal. The third column is the total processor time used by the process, in hours, minutes, and seconds. The last column is the name of the running program. In this example, bash is the name of the shell that interprets the commands I type in the terminal, emacs is my text editor, and ps is the program generating this output. By default, ps lists only the processes associated with the current terminal. If you use the -e flag, you get every process (including processes belonging to other users, which is a security flaw, in my opinion). On my system there are currently 233 processes. Here are some of them: PID 1 2 3 4 8 9 10 47 48 49 50 51 52 53 54 55

TTY ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?

TIME 00:00:17 00:00:00 00:00:02 00:00:00 00:00:00 00:00:00 00:00:16 00:00:00 00:00:00 00:00:00 00:00:00 00:00:00 00:00:00 00:00:00 00:00:00 00:00:00

CMD init kthreadd ksoftirqd/0 kworker/0:0 migration/0 rcu_bh rcu_sched cpuset khelper kdevtmpfs netns bdi-default kintegrityd kblockd ata_sff khubd

2.3.1

https://eng.libretexts.org/@go/page/40706

56 ? 57 ?

00:00:00 md 00:00:00 devfreq_wq

init is the first process created when the operating system starts. It creates many of the other processes, and then sits idle until the processes it created are done. kthreadd is a process the operating system uses to create new threads. We’ll talk more about threads later, but for now you can think of a thread as kind of a process. The k at the beginning stands for kernel, which is the part of the operating system responsible for core capabilities like creating threads. The extra d at the end stands for daemon, which is another name for processes like this that run in the background and provide operating system services. In this context, “daemon” is used in the sense of a helpful spirit, with no connotation of evil. Based on the name, you can infer that ksoftirqd is also a kernel daemon; specifically, it handles software interrupt requests, or “soft IRQ”. kworker is a worker process created by the kernel to do some kind of processing for the kernel. There are often multiple processes running these kernel services. On my system at the moment, there are 8 ksoftirqd processes and 35 kworker processes. I won’t go into more details about the other processes, but if you are interested you can search for more information about them. You should run ps on your system and compare your results to mine. This page titled 2.3: UNIX processes is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

2.3.2

https://eng.libretexts.org/@go/page/40706

CHAPTER OVERVIEW 3: Virtual memory 3.1: A bit of information theory 3.2: Memory and storage 3.3: Address spaces 3.6: Address translation 3.4: Memory segments 3.5: Static local variables

This page titled 3: Virtual memory is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1

3.1: A bit of information theory A bit is a binary digit; it is also a unit of information. If you have one bit, you can specify one of two possibilities, usually written 0 and 1. If you have two bits, there are 4 possible combinations, 00, 01, 10, and 11. In general, if you have b bits, you can indicate one of 2 values. A byte is 8 bits, so it can hold one of 256 values. b

Going in the other direction, suppose you want to store a letter of the alphabet. There are 26 letters, so how many bits do you need? With 4 bits, you can specify one of 16 values, so that’s not enough. With 5 bits, you can specify up to 32 values, so that’s enough for all the letters, with a few values left over. In general, if you want to specify one of N values, you should choose the smallest value of b so that 2 of both sides yields b ≤ log N .

b

≤N

. Taking the log base 2

2

Suppose I flip a coin and tell you the outcome. I have given you one bit of information. If I roll a six-sided die and tell you the outcome, I have given you log 6 bits of information. And in general, if the probability of the outcome is 1 in N , then the outcome contains log N bits of information. 2

2

Equivalently, if the probability of the outcome is p, then the information content is − log p . This quantity is called the selfinformation of the outcome. It measures how surprising the outcome is, which is why it is also called surprisal. If your horse has only one chance in 16 of winning, and he wins, you get 4 bits of information (along with the payout). But if the favorite wins 75% of the time, the news of the win contains only 0.42 bits. 2

Intuitively, unexpected news carries a lot of information; conversely, if there is something you were already confident of, confirming it contributes only a small amount of information. For several topics in this book, we will need to be comfortable converting back and forth between the number of bits, b , and the number of values they can encode, N = 2 . b

This page titled 3.1: A bit of information theory is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

3.1.1

https://eng.libretexts.org/@go/page/40718

3.2: Memory and storage While a process is running, most of its data is held in main memory, which is usually some kind of random access memory (RAM). On most current computers, main memory is volatile, which means that when the computer shuts down, the contents of main memory are lost. A typical desktop computer has 2–8 GiB of memory. GiB stands for “gibibyte,” which is 2 bytes. 30

If the process reads and writes files, those files are usually stored on a hard disk drive (HDD) or solid state drive (SSD). These storage devices are non-volatile, so they are used for long-term storage. Currently a typical desktop computer has a HDD with a capacity of 500 GB to 2 TB. GB stands for “gigabyte,” which is 10 bytes. TB stands for “terabyte,” which is 10 bytes. 9

12

You might have noticed that I used the binary unit GiB for the size of main memory and the decimal units GB and TB for the size of the HDD. For historical and technical reasons, memory is measured in binary units, and disk drives are measured in decimal units. In this book I will be careful to distinguish binary and decimal units, but you should be aware that the word “gigabyte” and the abbreviation GB are often used ambiguously. In casual use, the term “memory” is sometimes used for HDDs and SSDs as well as RAM, but the properties of these devices are very different, so we will need to distinguish them. I will use storage to refer to HDDs and SSDs. This page titled 3.2: Memory and storage is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

3.2.1

https://eng.libretexts.org/@go/page/40719

3.3: Address spaces Each byte in main memory is specified by an integer physical address. The set of valid physical addresses is called the physical address space. It usually runs from 0 to N − 1 , where N is the size of main memory. On a system with 1 GiB of physical memory, the highest valid address is 2 − 1 , which is 1,073,741,823 in decimal, or 0x3fff ffff in hexadecimal (the prefix 0x indicates a hexadecimal number). 30

However, most operating systems provide virtual memory, which means that programs never deal with physical addresses, and don’t have to know how much physical memory is available. Instead, programs work with virtual addresses, which are numbered from 0 to M − 1 , where M is the number of valid virtual addresses. The size of the virtual address space is determined by the operating system and the hardware it runs on. You have probably heard people talk about 32-bit and 64-bit systems. These terms indicate the size of the registers, which is usually also the size of a virtual address. On a 32-bit system, virtual addresses are 32 bits, which means that the virtual address space runs from 0 to 0xffff ffff. The size of this address space is 2 bytes, or 4 GiB. 32

On a 64-bit system, the size of the virtual address space is 2 bytes, or 2 ⋅ 1024 bytes. That’s 16 exbibytes, which is about a billion times bigger than current physical memories. It might seem strange that a virtual address space can be so much bigger than physical memory, but we will see soon how that works. 64

4

6

When a program reads and writes values in memory, it generates virtual addresses. The hardware, with help from the operating system, translates to physical addresses before accessing main memory. This translation is done on a per-process basis, so even if two processes generate the same virtual address, they would map to different locations in physical memory. Thus, virtual memory is one important way the operating system isolates processes from each other. In general, a process cannot access data belonging to another process, because there is no virtual address it can generate that maps to physical memory allocated to another process. This page titled 3.3: Address spaces is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

3.3.1

https://eng.libretexts.org/@go/page/40720

3.6: Address translation How does a virtual address (VA) get translated to a physical address (PA)? The basic mechanism is simple, but a simple implementation would be too slow and take too much space. So actual implementations are a bit more complicated.

Figure 3.6.1 : Diagram of the address translation process.

Most processors provide a memory management unit (MMU) that sits between the CPU and main memory. The MMU performs fast translation between VAs and PAs. 1. When a program reads or writes a variable, the CPU generates a VA. 2. The MMU splits the VA into two parts, called the page number and the offset. A “page” is a chunk of memory; the size of a page depends on the operating system and the hardware, but common sizes are 1–4 KiB. 3. The MMU looks up the page number in the translation lookaside buffer (TLB) and gets the corresponding physical page number. Then it combines the physical page number with the offset to produce a PA. 4. The PA is passed to main memory, which reads or writes the given location. The TLB contains cached copies of data from the page table (which is stored in kernel memory). The page table contains the mapping from virtual page numbers to physical page numbers. Since each process has its own page table, the TLB has to make sure it only uses entries from the page table of the process that’s running. Figure 3.6.1 shows a diagram of this process. To see how it all works, suppose that the VA is 32 bits and the physical memory is 1 GiB, divided into 1 KiB pages. Since 1 GiB is 2 bytes and 1 KiB is 2 bytes, there are 2 physical pages, sometimes called “frames.” The size of the virtual address space is 2 B and the size of a page is 2 B, so there are 2 virtual pages. The size of the offset is determined by the page size. In this example the page size is 2 B, so it takes 10 bits to specify a byte on a page. If a VA is 32 bits and the offset is 10 bits, the remaining 22 bits make up the virtual page number. Since there are 2 physical pages, each physical page number is 20 bits. Adding in the 10 bit offset, the resulting PAs are 30 bits. 30

10

20

32

10

22

10

20

So far this all seems feasible. But let’s think about how big a page table might have to be. The simplest implementation of a page table is an array with one entry for each virtual page. Each entry would contain a physical page number, which is 20 bits in this example, plus some additional information about each frame. So we expect 3–4 bytes per entry. But with 222 virtual pages, the page table would require 224 bytes, or 16 MiB. And since we need a page table for each process, a system running 256 processes would need 2 bytes, or 4 GiB, just for page tables! And that’s just with 32-bit virtual addresses. With 48- or 64-bit VAs, the numbers are ridiculous. 32

Fortunately, we don’t actually need that much space, because most processes don’t use even a small fraction of their virtual address space. And if a process doesn’t use a virtual page, we don’t need an entry in the page table for it. Another way to say the same thing is that page tables are “sparse”, which implies that the simple implementation, an array of page table entries, is a bad idea. Fortunately, there are several good implementations for sparse arrays. One option is a multilevel page table, which is what many operating systems, including Linux, use. Another option is an associative table, where each entry includes both the virtual page number and the physical page number. Searching an associative table can be slow in software, but in hardware we can search the entire table in parallel, so associative arrays are often used to represent the page table entries in the TLB. You can read more about these implementations at http://en.Wikipedia.org/wiki/Page_table; you might find the details interesting. But the fundamental idea is that page tables are sparse, so we have to choose a good implementation for sparse arrays. I mentioned earlier that the operating system can interrupt a running process, save its state, and then run another process. This mechanism is called a context switch. Since each process has its own page table, the operating system has to work with the MMU

3.6.1

https://eng.libretexts.org/@go/page/40723

to make sure each process gets the right page table. In older machines, the page table information in the MMU had to be replaced during every context switch, which was expensive. In newer systems, each page table entry in the MMU includes the process ID, so page tables from multiple processes can be in the MMU at the same time. This page titled 3.6: Address translation is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

3.6.2

https://eng.libretexts.org/@go/page/40723

3.4: Memory segments The data of a running process is organized into five segments: The code segment contains the program text; that is, the machine language instructions that make up the program. The static segment contains immutable values, like string literals. For example, if your program contains the string "Hello, World" , those characters will be stored in the static segment. The global segment contains global variables and local variables that are declared static . The heap segment contains chunks of memory allocated at run time, most often by calling the C library function malloc . The stack segment contains the call stack, which is a sequence of stack frames. Each time a function is called, a stack frame is allocated to contain the parameters and local variables of the function. When the function completes, its stack frame is removed from the stack. The arrangement of these segments is determined partly by the compiler and partly by the operating system. The details vary from one system to another, but in the most common arrangement: The text segment is near the “bottom” of memory, that is, at addresses near 0. The static segment is often just above the text segment, that is, at higher addresses. The global segment is often just above the static segment. The heap is often above the global segment. As it expands, it grows up toward larger addresses. The stack is near the top of memory; that is, near the highest addresses in the virtual address space. As the stack expands, it grows down toward smaller addresses. To determine the layout of these segments on your system, try running this program, which is in aspace.c in the repository for this book (see Section 0.2). #include #include int global; int main () { int local = 5; void *p = malloc(128); char *s = "Hello, World"; printf printf printf printf printf

("Address of main is %p\n", main); ("Address of global is %p\n", &global); ("Address of local is %p\n", &local); ("p points to %p\n", p); ("s points to %p\n", s);

} main is the name of a function; when it is used as a variable, it refers to the address of the first machine language instruction in main , which we expect to be in the text segment. global is a global variable, so we expect it to be in the global segment. local is a local variable, so we expect it to be on the stack. s refers to a “string literal", which is a string that appears as part of the program (as opposed to a string that is read from a file, input by a user, etc.). We expect the location of the string to be in the static segment (as opposed to the pointer, s , which is a local variable).

3.4.1

https://eng.libretexts.org/@go/page/40721

p contains an address returned by malloc , which allocates space in the heap. “malloc” stands for “memory allocate.” The format sequence %p tells printf to format each address as a “pointer”, so it displays the results in hexadecimal. When I run this program, the output looks like this (I added spaces to make it easier to read): Address of main is Address of global is Address of local is p points to s points to

0x 40057d 0x 60104c 0x7ffe6085443c 0x 16c3010 0x 4006a4

As expected, the address of main is the lowest, followed by the location of the string literal. The location of global is next, then the address p points to. The address of local is much bigger. The largest address has 12 hexadecimal digits. Each hex digit corresponds to 4 bits, so it is a 48-bit address. That suggests that the usable part of the virtual address space is 2 bytes. 48

As an exercise, run this program on your computer and compare your results to mine. Add a second call to malloc and check whether the heap on your system grows up (toward larger addresses). Add a function that prints the address of a local variable, and check whether the stack grows down. This page titled 3.4: Memory segments is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

3.4.2

https://eng.libretexts.org/@go/page/40721

3.5: Static local variables Local variables on the stack are sometimes called automatic, because they are allocated automatically when a function is called, and freed automatically when the function returns. In C there is another kind of local variable, called static, which is allocated in the global segment. It is initialized when the program starts and keeps its value from one function call to the next. For example, the following function keeps track of how many times it has been called. int times_called() { static int counter = 0; counter++; return counter; } The keyword static indicates that counter is a static local variable. The initialization happens only once, when the program starts. If you add this function to aspace.c you can confirm that counter is allocated in the global segment along with global variables, not in the stack. This page titled 3.5: Static local variables is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

3.5.1

https://eng.libretexts.org/@go/page/40722

CHAPTER OVERVIEW 4: Files and file systems 4.1: Files and file systems 4.2: Disk performance 4.3: Disk metadata 4.4: Block allocation 4.5: Everything is a file?

This page titled 4: Files and file systems is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1

4.1: Files and file systems When a process completes (or crashes), any data stored in main memory is lost. But data stored on a hard disk drive (HDD) or solid state drive (SSD) is “persistent;” that is, it survives after the process completes, even if the computer shuts down. Hard disk drives are complicated. Data is stored in blocks, which are laid out in sectors, which make up tracks, which are arranged in concentric circles on platters. Solid state drives are simpler in one sense, because blocks are numbered sequentially, but they raise a different complication: each block can be written a limited number of times before it becomes unreliable. As a programmer, you don’t want to deal with these complications. What you want is an appropriate abstraction of persistent storage hardware. The most common abstraction is called a “file system.” Abstractly: A “file system” is a mapping from each file’s name to its contents. If you think of the names as keys, and the contents as values, a file system is a kind of key-value database (see https://en.Wikipedia.org/wiki/Key-value_database). A “file” is a sequence of bytes. File names are usually strings, and they are usually “hierarchical”; that is, the string specifies a path from a top-level directory (or folder), through a series of subdirectories, to a specific file. The primary difference between the abstraction and the underlying mechanism is that files are byte-based and persistent storage is block-based. The operating system translates byte-based file operations in the C library into block-based operations on storage devices. Typical block sizes are 1–8 KiB. For example, the following code opens a file and reads the first byte: FILE *fp = fopen("/home/downey/file.txt", "r"); char c = fgetc(fp); fclose(fp); When this code runs: 1. fopen uses the filename to find the top-level directory, called / , the subdirectory home , and the sub-subdirectory downey . 2. It finds the file named file.txt and “opens” it for reading, which means it creates a data structure that represents the file being read. Among other things, this data structure keeps track of how much of the file has been read, called the “file position”. In DOS, this data structure is called a File Control Block, but I want to avoid that term because in UNIX it means something else. In UNIX, there seems to be no good name for it. It is an entry in the open file table, so I will call it an OpenFileTableEntry. 3. When we call fgetc , the operating system checks whether the next character of the file is already in memory. If so, it reads the next character, advances the file position, and returns the result. 4. If the next character is not in memory, the operating system issues an I/O request to get the next block. Disk drives are slow, so a process waiting for a block from disk is usually interrupted so another process can run until the data arrives. 5. When the I/O operation is complete, the new block of data is stored in memory, and the process resumes. It reads the first character and stores it as a local variable. 6. When the process closes the file, the operating system completes or cancels any pending operations, removes data stored in memory, and frees the OpenFileTableEntry. The process for writing a file is similar, but there are some additional steps. Here is an example that opens a file for writing and changes the first character. FILE *fp = fopen("/home/downey/file.txt", "w"); fputc('b', fp); fclose(fp);

4.1.1

https://eng.libretexts.org/@go/page/42857

When this code runs: 1. Again, fopen uses the filename to find the file. If it does not already exist, it creates a new file and adds an entry in the parent directory, /home/downey . 2. The operating system creates an OpenFileTableEntry that indicates that the file is open for writing, and sets the file position to 0. 3. fputc attempts to write (or re-write) the first byte of the file. If the file already exists, the operating system has to load the first block into memory. Otherwise it allocates a new block in memory and requests a new block on disk. 4. After the block in memory is modified, it might not be copied back to the disk right away. In general, data written to a file is “buffered”, which means it is stored in memory and only written to disk when there is at least one block to write. 5. When the file is closed, any buffered data is written to disk and the OpenFileTableEntry is freed. To summarize, the C library provides the abstraction of a file system that maps from file names to streams of bytes. This abstraction is built on top of storage devices that are actually organized in blocks. This page titled 4.1: Files and file systems is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

4.1.2

https://eng.libretexts.org/@go/page/42857

4.2: Disk performance I mentioned earlier that disk drives are slow. On current HDDs, the average time to read a block from disk to memory might be 5– 25 ms (see https://en.Wikipedia.org/wiki/Hard_disk_drive_performance_characteristics). SSDs are faster, taking 25 µs to read a 4 KiB block and 250 µs to write one (see http://en.Wikipedia.org/wiki/Ssd#Controller). To put these numbers in perspective, let’s compare them to the clock cycle of the CPU. A processor with clock rate 2 GHz completes one clock cycle every 0.5 ns. The time to get a byte from memory to the CPU is typically around 100 ns. If the processor completes one instruction per clock cycle, it would complete 200 instructions while waiting for a byte from memory. In one microsecond, it would complete 2000 instructions, so while waiting 25 µs for a byte from an SSD, it would complete 50,000. In one millisecond, it would complete 2,000,000 instructions, so while waiting 20 ms for a byte from a HDD, it might complete 40 million. If there’s nothing for the CPU to do while it waits, it would be idle. That’s why the operating system generally switches to another process while it is waiting for data from disk. The gap in performance between main memory and persistent storage is one of the major challenges of computer system design. Operating systems and hardware provide several features intended to “fill in” this gap: Block transfers: The time it takes to load a single byte from disk is 5–25 ms. By comparison, the additional time to load an 8 KiB block is negligible. So systems generally try to read large blocks each time they access the disk. Prefetching: Sometimes the operating system can predict that a process will read a block and start loading it before it is requested. For example, if you open a file and read the first block, there is a good chance you will go on to read the second block. The operating system might start loading additional blocks before they are requested. Buffering: As I mentioned, when you write a file, the operating system stores the data in memory and only writes it to disk later. If you modify the block several times while it is in memory, the system only has to write it to disk once. Caching: If a process has used a block recently, it is likely to use it again soon. If the operating system keeps a copy of the block in memory, it can handle future requests at memory speed. Some of these features are also implemented in hardware. For example, some disk drives provide a cache that stores recently-used blocks, and many disk drives read more than one block at a time, even if only one is requested. These mechanisms generally improve the performance of programs, but they don’t change the behavior. Usually programmers don’t have to think about them, with two exceptions: (1) if the performance of a program is unexpectedly bad, you might have to know something about these mechanisms to diagnose the problem, and (2) when data is buffered, it can be harder to debug a program. For example, if a program prints a value and then crashes, the value might not appear, because it might be in a buffer. Similarly, if a program writes data to disk and then the computer loses power, the data might be lost if it is in a cache and not yet on disk. This page titled 4.2: Disk performance is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

4.2.1

https://eng.libretexts.org/@go/page/40725

4.3: Disk metadata The blocks that make up a file might be arranged contiguously on disk, and file system performance is generally better if they are, but most operating systems don’t require contiguous allocation. They are free to place a block anywhere on disk, and they use various data structures to keep track of them. In many UNIX file systems, that data structure is called an “inode,” which stands for “index node”. More generally, information about files, including the location of their blocks, is called “metadata”. (The content of the file is data, so information about the file is data about data, hence “meta”.) Since inodes reside on disk along with the rest of the data, they are designed to fit neatly into disk blocks. A UNIX inode contains information about a file, including the user ID of the file owner; permission flags indicating who is allowed to read, write, or execute it; and timestamps that indicate when it was last modified and accessed. In addition, it contains block numbers for the first 12 blocks that make up the file. If the block size is 8 KiB, the first 12 blocks make up 96 KiB. On most systems, that’s big enough for a large majority of files, but it’s definitely not big enough for all of them. That’s why the inode also contains a pointer to an “indirection block”, which contains nothing but pointers to other blocks. The number of pointers in an indirection block depends on the sizes of the blocks and the block numbers, but it is often 1024. With 1024 block numbers and 8 KiB blocks, an indirection block can address 8 MiB. That’s big enough for all but the largest files, but still not big enough for all. That’s why the inode also contains a pointer to a “double indirection block”, which contains pointers to indirection blocks. With 1024 indirection blocks, we can address 8 GiB. And if that’s not big enough, there is (finally) a triple indirection block, which contains pointers to double indirection blocks, yielding a maximum file size of 8 TiB. When UNIX inodes were designed, that seemed big enough to serve for a long time. But that was a long time ago. As an alternative to indirection blocks, some files systems, like FAT, use a File Allocation Table that contains one entry for each block, called a “cluster” in this context. A root directory contains a pointer to the first cluster in each file. The FAT entry for each cluster points to the next cluster in the file, similar to a linked list. For more details, see http://en.Wikipedia.org/wiki/File_Allocation_Table. This page titled 4.3: Disk metadata is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

4.3.1

https://eng.libretexts.org/@go/page/40726

4.4: Block allocation File systems have to keep track of which blocks belong to each file; they also have to keep track of which blocks are available for use. When a new file is created, the file system finds an available block and allocates it. When a file is deleted, the file system makes its blocks available for re-allocation. The goals of the block allocation system are: Speed: Allocating and freeing blocks should be fast. Minimal space overhead: The data structures used by the allocator should be small, leaving as much space as possible for data. Minimal fragmentation: If some blocks are left unused, or some are only partially used, the unused space is called “fragmentation”. Maximum contiguity: Data that is likely to be used at the same time should be physically contiguous, if possible, to improve performance. It is hard to design a file system that achieves all of these goals, especially since file system performance depends on “workload characteristics” like file sizes, access patterns, etc. A file system that is well tuned for one workload might not perform as well for another. For this reason, most operating systems support several kinds of file systems, and file system design is an active area of research and development. In the last decade, Linux systems have migrated from ext2, which was a conventional UNIX file system, to ext3, a “journaling” file system intended to improve speed and contiguity, and more recently to ext4, which can handle larger files and file systems. Within the next few years, there might be another migration to the B-tree file system, Btrfs. This page titled 4.4: Block allocation is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

4.4.1

https://eng.libretexts.org/@go/page/40727

4.5: Everything is a file? The file abstraction is really a “stream of bytes” abstraction, which turns out to be useful for many things, not just file systems. One example is the UNIX pipe, which is a simple form of inter-process communication. Processes can be set up so that output from one process is taken as input into another process. For the first process, the pipe behaves like a file open for writing, so it can use C library functions like fputs and fprintf . For the second process, the pipe behaves like a file open for reading, so it uses fgets and fscanf . Network communication also uses the stream of bytes abstraction. A UNIX socket is a data structure that represents a communication channel between processes on different computers (usually). Again, processes can read data from and write data to a socket using “file” handling functions. Reusing the file abstraction makes life easier for programmers, since they only have to learn one API (application program interface). It also makes programs more versatile, since a program intended to work with files can also work with data coming from pipes and other sources. This page titled 4.5: Everything is a file? is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

4.5.1

https://eng.libretexts.org/@go/page/40728

CHAPTER OVERVIEW 5: More bits and bytes 5.1: Representing integers 5.2: Bitwise operators 5.3: Representing floating-point numbers 5.4: Unions and memory errors 5.5: Representing strings

This page titled 5: More bits and bytes is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1

5.1: Representing integers You probably know that computers represent numbers in base 2, also known as binary. For positive numbers, the binary representation is straightforward; for example, the representation for 5 is b101. 10

For negative numbers, the most obvious representation uses a sign bit to indicate whether a number is positive or negative. But there is another representation, called “two’s complement” that is much more common because it is easier to work with in hardware. To find the two’s complement of a negative number, −x, find the binary representation of x, flip all the bits, and add 1. For example, to represent −5 , start with the representation of 5 , which is b0000 0101 if we write the 8-bit version. Flipping all the bits and adding 1 yields b1111 1011. 10

10

In two’s complement, the leftmost bit acts like a sign bit; it is 0 for positive numbers and 1 for negative numbers. To convert from an 8-bit number to 16-bits, we have to add more 0’s for a positive number and add 1’s for a negative number. In effect, we have to copy the sign bit into the new bits. This process is called “sign extension”. In C all integer types are signed (able to represent positive and negative numbers) unless you declare them unsigned . The difference, and the reason this declaration is important, is that operations on unsigned integers don’t use sign extension. This page titled 5.1: Representing integers is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

5.1.1

https://eng.libretexts.org/@go/page/40732

5.2: Bitwise operators People learning C are sometimes confused about the bitwise operators & and | . These operators treat integers as bit vectors and compute logical operations on corresponding bits. For example, & computes the AND operation, which yields 1 if both operands are 1, and 0 otherwise. Here is an example of & applied to two 4-bit numbers: 1100 & 1010 ---1000 In C, this means that the expression 12 & 10 has the value 8. Similarly, | computes the OR operation, which yields 1 if either operand is 1, and 0 otherwise. 1100 | 1010 ---1110 So the expression 12 | 10 has the value 14. Finally, ^ computes the XOR operation, which yields 1 if either operand is 1, but not both. 1100 ^ 1010 ---0110 So the expression 12 ^ 10 has the value 6. Most commonly, & is used to clear a set of bits from a bit vector, | is used to set bits, and ^ is used to flip, or “toggle” bits. Here are the details: Clearing bits: For any value x, x & 0 is 0, and x & 1 is x. So if you AND a vector with 3, it selects only the two rightmost bits, and sets the rest to 0. xxxx & 0011 ---00xx In this context, the value 3 is called a “mask” because it selects some bits and masks the rest. Setting bits: Similarly, for any x, x | 0 is x, and x | 1 is 1. So if you OR a vector with 3, it sets the rightmost bits, and leaves the rest alone: xxxx | 0011 ---xx11

5.2.1

https://eng.libretexts.org/@go/page/40733

Toggling bits: Finally, if you XOR a vector with 3, it flips the rightmost bits and leaves the rest alone. As an exercise, see if you can compute the two’s complement of 12 using ^ . Hint: what’s the two’s complement representation of -1? C also provides shift operators, > , which shift bits left and right. Each left shift doubles a number, so 5 1 is 2 and 2 >> 1 is 1. This page titled 5.2: Bitwise operators is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

5.2.2

https://eng.libretexts.org/@go/page/40733

5.3: Representing floating-point numbers Floating-point numbers are represented using the binary version of scientific notation. In decimal notation, large numbers are written as the product of a coefficient and 10 raised to an exponent. For example, the speed of light in m/s is approximately 2.998 ⋅ 10 . 8

Most computers use the IEEE standard for floating-point arithmetic. The C type float usually corresponds to the 32-bit IEEE standard; double usually corresponds to the 64-bit standard. In the 32-bit standard, the leftmost bit is the sign bit, s . The next 8 bits are the exponent, q, and the last 23 bits are the coefficient, c . The value of a floating-point number is s

(−1 )

q

⋅2

Well, that’s almost correct, but there’s one more wrinkle. Floating-point numbers are usually normalized so that there is one digit before the point. For example, in base 10, we prefer 2.998 ⋅ 10 rather than 2998 ⋅ 10 or any other equivalent expression. In base 2, a normalized number always has the digit 1 before the binary point. Since the digit in this location is always 1, we can save space by leaving it out of the representation. 8

5

For example, the integer representation of 13 is b1101. In floating point, that’s 1.101 ⋅ 2 , so the exponent is 3 and the part of the coefficient that would be stored is 101 (followed by 20 zeros). 3

10

Well, that’s almost correct, but there’s one more wrinkle. The exponent is stored with a “bias”. In the 32-bit standard, the bias is 127, so the exponent 3 would be stored as 130. To pack and unpack floating-point numbers in C, we can use a union and bitwise operations. Here’s an example: union { float f; unsigned int u; } p; p.f = -13.0; unsigned int sign = (p.u >> 31) & 1; unsigned int exp = (p.u >> 23) & 0xff; unsigned int coef_mask = (1 counter); shared->counter++; } When child_code returns, entry invokes pthread_exit which can be used to pass a value to the thread that joins with this thread. In this case, the child has nothing to say, so we pass NULL . Finally, here is the code that creates the child threads: int i; pthread_t child[NUM_CHILDREN]; Shared *shared = make_shared(1000000); for (i=0; icounter); shared->counter++; mutex_unlock(shared->mutex); } Before any thread can access counter , it has to “lock” the mutex, which has the effect of barring all other threads. Suppose Thread A has locked the mutex and is in the middle of child_code . If Thread B arrives and executes mutex_lock , it blocks. When Thread A is done, it executes mutex_unlock , which allows Thread B to proceed. In effect, the threads line up to execute child_code one at a time, so they can’t interfere with each other. When I run this code with 5 children, I get:

9.4.1

https://eng.libretexts.org/@go/page/40763

counter counter counter counter counter

= = = = =

0 1 2 3 4

And that satisfies the requirements. In order for this solution to work, I have to add the Mutex to the Shared struct: typedef struct { int counter; Mutex *mutex; } Shared; And initialize it in make_shared Shared *make_shared(int end) { Shared *shared = check_malloc(sizeof(Shared)); shared->counter = 0; shared->mutex = make_mutex(); //-- this line is new return shared; } The code in this section is in counter_mutex.c . The definition of Mutex is in mutex.c , which I explain in the next section. This page titled 9.4: Synchronization errors is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

9.4.2

https://eng.libretexts.org/@go/page/40763

9.5: Mutex My definition of Mutex is a wrapper for a type called pthread_mutex_t , which is defined in the POSIX threads API. To create a POSIX mutex, you have to allocate space for a pthread_mutex_init .

pthread_mutex_t

type and then call

One of the problems with this API is that pthread_mutex_t behaves like a structure, so if you pass it as an argument, it makes a copy, which makes the mutex behave incorrectly. To avoid that, you have to pass pthread_mutex_t by address. My code makes it easier to get that right. It defines a type, pthread_mutex_t :

Mutex , which is just a more readable name for

#include typedef pthread_mutex_t Mutex; Then it defines make_mutex , which allocates space and initializes the mutex: Mutex *make_mutex() { Mutex *mutex = check_malloc(sizeof(Mutex)); int n = pthread_mutex_init(mutex, NULL); if (n != 0) perror_exit("make_lock failed"); return mutex; } The return value is a pointer, which you can pass around as an argument without causing unwanted copying. The functions to lock and unlock the mutex are simple wrappers for POSIX functions: void mutex_lock(Mutex *mutex) { int n = pthread_mutex_lock(mutex); if (n != 0) perror_exit("lock failed"); } void mutex_unlock(Mutex *mutex) { int n = pthread_mutex_unlock(mutex); if (n != 0) perror_exit("unlock failed"); } This code is in mutex.c and the header file mutex.h . This page titled 9.5: Mutex is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

9.5.1

https://eng.libretexts.org/@go/page/40764

CHAPTER OVERVIEW 10: Condition variables Many simple synchronization problems can be solved using mutexes as shown in the previous chapter. In this chapter I introduce a bigger challenge, the well-known “Producer-Consumer problem”, and a new tool to solve it, the condition variable. 10.1: The work queue 10.2: Producers and consumers 10.3: Mutual exclusion 10.4: Condition variables 10.5: Condition variable implementation

This page titled 10: Condition variables is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1

10.1: The work queue In some multi-threaded programs, threads are organized to perform different tasks. Often they communicate with each other using a queue, where some threads, called “producers”, put data into the queue and other threads, called “consumers”, take data out. For example, in applications with a graphical user interface, there might be one thread that runs the GUI, responding to user events, and another thread that processes user requests. In that case, the GUI thread might put requests into a queue and the “back end” thread might take requests out and process them. To support this organization, we need a queue implementation that is “thread safe”, which means that both threads (or more than two) can access the queue at the same time. And we need to handle the special cases when the queue is empty and, if the size of the queue is bounded, when the queue is full. I’ll start with a simple queue that is not thread safe, then we’ll see what goes wrong and fix it. The code for this example is in the repository for this book, in a folder called queue. The file queue.c contains a basic implementation of a circular buffer, which you can read about at https://en.Wikipedia.org/wiki/Circular_buffer. Here’s the structure definition: typedef struct { int *array; int length; int next_in; int next_out; } Queue; array is the array that contains the elements of the queue. For this example the elements are ints, but more generally they would be structures that contain user events, items of work, etc. length is the length of the array. next_in is an index into the array that indices where the next element should be added; similarly, next_out is the index of the next element that should be removed. make_queue allocates space for this structure and initializes the fields: Queue *make_queue(int length) { Queue *queue = (Queue *) malloc(sizeof(Queue)); queue->length = length + 1; queue->array = (int *) malloc(length * sizeof(int)); queue->next_in = 0; queue->next_out = 0; return queue; } The initial value for next_out needs some explaining. Since the queue is initially empty, there is no next element to remove, so next_out is invalid. Setting next_out == next_in is a special case that indicates that the queue is empty, so we can write: int queue_empty(Queue *queue) { return (queue->next_in == queue->next_out); } Now we can add elements to the queue using queue_push :

10.1.1

https://eng.libretexts.org/@go/page/40634

void queue_push(Queue *queue, int item) { if (queue_full(queue)) { perror_exit("queue is full"); } queue->array[queue->next_in] = item; queue->next_in = queue_incr(queue, queue->next_in); } If the queue is full, queue_push prints an error message and exits. I will explain queue_full soon. If the queue is not full, queue_push inserts the new element and then increments next_in using queue_incr : int queue_incr(Queue *queue, int i) { return (i+1) % queue->length; } When the index, i , gets to the end of the array, it wraps around to 0. And that’s where we run into a tricky part. If we keep adding elements to the queue, eventually next_in wraps around and catches up with next_out . But if next_in == next_out , we would incorrectly conclude that the queue was empty. To avoid that, we define another special case to indicate that the queue is full: int queue_full(Queue *queue) { return (queue_incr(queue, queue->next_in) == queue->next_out); } If incrementing next_in lands on next_out , that means we can’t add another element without making the queue seem empty. So we stop one element before the “end” (keeping in mind that the end of the queue can be anywhere, not necessarily the end of the array). Now we can write queue_pop , which removes and returns the next element from the queue: int queue_pop(Queue *queue) { if (queue_empty(queue)) { perror_exit("queue is empty"); } int item = queue->array[queue->next_out]; queue->next_out = queue_incr(queue, queue->next_out); return item; } If you try to pop from an empty queue, queue_pop prints an error message and exits. This page titled 10.1: The work queue is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

10.1.2

https://eng.libretexts.org/@go/page/40634

10.2: Producers and consumers Now let’s make some threads to access this queue. Here’s the producer code: void *producer_entry(void *arg) { Shared *shared = (Shared *) arg; for (int i=0; iqueue, i); } pthread_exit(NULL); } Here’s the consumer code: void *consumer_entry(void *arg) { int item; Shared *shared = (Shared *) arg; for (int i=0; iqueue); printf("consuming item %d\n", item); } pthread_exit(NULL); } Here’s the parent code that starts the threads and waits for them pthread_t child[NUM_CHILDREN]; Shared *shared = make_shared(); child[0] = make_thread(producer_entry, shared); child[1] = make_thread(consumer_entry, shared); for (int i=0; iqueue = make_queue(QUEUE_LENGTH); return shared; } The code we have so far is a good starting place, but it has several problems: Access to the queue is not thread safe. Different threads could access array , next_in , and next_out at the same time and leave the queue in a broken, “inconsistent” state. If the consumer is scheduled first, it finds the queue empty, print an error message, and exits. We would rather have the consumer block until the queue is not empty. Similarly, we would like the producer to block if the queue is full. In the next section, we solve the first problem with a Mutex . In the following section, we solve the second problem with condition variables. This page titled 10.2: Producers and consumers is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

10.2.2

https://eng.libretexts.org/@go/page/40635

10.3: Mutual exclusion We can make the queue thread safe with a mutex. This version of the code is in queue_mutex.c . First we add a Mutex pointer to the queue structure: typedef struct { int *array; int length; int next_in; int next_out; Mutex *mutex; } Queue;

//-- this line is new

And initialize the Mutex in make_queue : Queue *make_queue(int length) { Queue *queue = (Queue *) malloc(sizeof(Queue)); queue->length = length; queue->array = (int *) malloc(length * sizeof(int)); queue->next_in = 0; queue->next_out = 0; queue->mutex = make_mutex(); //-- new return queue; } Next we add synchronization code to queue_push : void queue_push(Queue *queue, int item) { mutex_lock(queue->mutex); //-- new if (queue_full(queue)) { mutex_unlock(queue->mutex); //-- new perror_exit("queue is full"); } queue->array[queue->next_in] = item; queue->next_in = queue_incr(queue, queue->next_in); mutex_unlock(queue->mutex); //-- new } Before checking whether the queue is full, we have to lock the Mutex . If the queue is full, we have to unlock the Mutex before exiting; otherwise the thread would leave it locked and no other threads could proceed. The synchronization code for queue_pop is similar: int queue_pop(Queue *queue) { mutex_lock(queue->mutex); if (queue_empty(queue)) { mutex_unlock(queue->mutex);

10.3.1

https://eng.libretexts.org/@go/page/40636

perror_exit("queue is empty"); } int item = queue->array[queue->next_out]; queue->next_out = queue_incr(queue, queue->next_out); mutex_unlock(queue->mutex); return item; } Note that the other Queue functions, queue_full , queue_empty , and queue_incr do not try to lock the mutex. Any thread that calls these functions is required to lock the mutex first; this requirement is part of the documented interface for these functions. With this additional code, the queue is thread safe; if you run it, you should not see any synchronization errors. But it is likely that the consumer will exit at some point because the queue is empty, or the producer will exit because the queue is full, or both. The next step is to add condition variables. This page titled 10.3: Mutual exclusion is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

10.3.2

https://eng.libretexts.org/@go/page/40636

10.4: Condition variables A condition variable is a data structure associated with a condition; it allows threads to block until the condition becomes true. For example, thread_pop might want check whether the queue is empty and, if so, wait for a condition like “queue not empty”. Similarly, thread_push might want to check whether the queue is full and, if so, block until it is not full. I’ll handle the first condition here, and you will have a chance to handle the second condition as an exercise. First we add a condition variable to the Queue structure: typedef struct { int *array; int length; int next_in; int next_out; Mutex *mutex; Cond *nonempty; } Queue;

//-- new

And initialize it in make_queue : Queue *make_queue(int length) { Queue *queue = (Queue *) malloc(sizeof(Queue)); queue->length = length; queue->array = (int *) malloc(length * sizeof(int)); queue->next_in = 0; queue->next_out = 0; queue->mutex = make_mutex(); queue->nonempty = make_cond(); //-- new return queue; } Now in queue_pop , if we find the queue empty, we don’t exit; instead we use the condition variable to block: int queue_pop(Queue *queue) { mutex_lock(queue->mutex); while (queue_empty(queue)) { cond_wait(queue->nonempty, queue->mutex); }

//-- new

int item = queue->array[queue->next_out]; queue->next_out = queue_incr(queue, queue->next_out); mutex_unlock(queue->mutex); cond_signal(queue->nonfull); //-- new return item; }

10.4.1

https://eng.libretexts.org/@go/page/40637

cond_wait is complicated, so let’s take it slow. The first argument is the condition variable; in this case, the condition we are waiting for is “queue not empty”. The second argument is the mutex that protects the queue. When the thread that locked the mutex calls cond_wait , it unlocks the mutex and then blocks. This is important. If cond_wait did not unlock the mutex before blocking, no other thread would be able to access the queue, no more items could be added, and the queue would always be empty. So while the consumer is blocked on nonempty , the producer can run. Let’s see what happens when the producer runs queue_push : void queue_push(Queue *queue, int item) { mutex_lock(queue->mutex); if (queue_full(queue)) { mutex_unlock(queue->mutex); perror_exit("queue is full"); } queue->array[queue->next_in] = item; queue->next_in = queue_incr(queue, queue->next_in); mutex_unlock(queue->mutex); cond_signal(queue->nonempty); //-- new } Just as before, queue_push locks the Mutex and checks whether the queue is full. Assuming it is not, queue_push adds a new element to the queue and then unlocks the Mutex . But before returning, it does one more thing: it “signals” the condition variable nonempty . Signalling a condition variable usually indicates that the condition is true. If there are no threads waiting on the condition variable, the signal has no effect. If there are threads waiting on the condition variable, one of them gets unblocked and resumes execution of cond_wait . But before the awakened thread can return from cond_wait , it has to wait for and lock the Mutex , again. Now go back to queue_pop and see what happens when the thread returns from cond_wait . It loops back to the top of the while loop and checks the condition again. I’ll explain why in just a second, but for now let’s assume that the condition is true; that is, the queue is not empty. When the consumer thread exits the while loop, we know two things: (1) the condition is true, so there is at least one item in the queue, and (2) the Mutex is locked, so it is safe to access the queue. After removing an item, queue_pop unlocks the mutex and returns. In the next section I’ll show you how my Cond code works, but first I want to answer two frequently-asked questions: Why is cond_wait inside a while loop rather than an if statement; that is, why do we have to check the condition again after returning from cond_wait ? The primary reason you have to re-check the condition is the possibility of an intercepted signal. Suppose Thread A is waiting on nonempty . Thread B adds an item to the queue and signals nonempty . Thread A wakes up an tries to lock the mutex, but before it gets the chance, Evil Thread C swoops in, locks the mutex, pops the item from the queue, and unlocks the mutex. Now the queue is empty again, but Thread A is not blocked any more. Thread A could lock the mutex and returns from cond_wait . If Thread A does not check the condition again, it would try to pop an element from an empty queue, and probably cause an error. The other question that comes up when people learn about condition variables is “How does the condition variable know what condition it is associated with?” This question is understandable because there is no explicit connection between a Cond structure and the condition it relates to. The connection is implicit in the way it is used.

10.4.2

https://eng.libretexts.org/@go/page/40637

Here’s one way to think of it: the condition associated with a Cond is the thing that is false when you call cond_wait and true when you call cond_signal . Because threads have to check the condition when they return from cond_wait , it is not strictly necessary to call cond_signal only when the condition is true. If you have reason to think the condition might be true, you could call cond_signal as a suggestion that now is a good time to check. This page titled 10.4: Condition variables is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

10.4.3

https://eng.libretexts.org/@go/page/40637

10.5: Condition variable implementation The Cond structure I used in the previous section is a wrapper for a type called pthread_cond_t , which is defined in the POSIX threads API. It is very similar to Mutex, which is a wrapper for pthread_mutex_t . Both wrappers are defined in utils.c and utils.h . Here’s the typedef : typedef pthread_cond_t Cond; make_cond allocates space, initializes the condition variable, and returns a pointer: Cond *make_cond() { Cond *cond = check_malloc(sizeof(Cond)); int n = pthread_cond_init(cond, NULL); if (n != 0) perror_exit("make_cond failed"); return cond; } And here are the wrappers for cond_wait and cond_signal . void cond_wait(Cond *cond, Mutex *mutex) { int n = pthread_cond_wait(cond, mutex); if (n != 0) perror_exit("cond_wait failed"); } void cond_signal(Cond *cond) { int n = pthread_cond_signal(cond); if (n != 0) perror_exit("cond_signal failed"); } At this point there should be nothing too surprising there. This page titled 10.5: Condition variable implementation is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

10.5.1

https://eng.libretexts.org/@go/page/40638

CHAPTER OVERVIEW 11: Semaphores in C Semaphores are a good way to learn about synchronization, but they are not as widely used, in practice, as mutexes and condition variables. Nevertheless, there are some synchronization problems that can be solved simply with semaphores, yielding solutions that are more demonstrably correct. This chapter presents a C API for working with semaphores and my code for making it easier to work with. And it presents a final challenge: can you write an implementation of a semaphore using mutexes and condition variables? The code for this chapter is in directory semaphore in the repository for this book (see Section 0.2). 11.1: POSIX Semaphores 11.2: Producers and consumers with semaphores 11.3: Make your own semaphores

This page titled 11: Semaphores in C is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

1

11.1: POSIX Semaphores A semaphore is a data structure used to help threads work together without interfering with each other. The POSIX standard specifies an interface for semaphores; it is not part of Pthreads, but most UNIXes that implement Pthreads also provide semaphores. POSIX semaphores have type sem_t . As usual, I put a wrapper around sem_t to make it easier to use. The interface is defined in sem.h : typedef sem_t Semaphore; Semaphore *make_semaphore(int value); void semaphore_wait(Semaphore *sem); void semaphore_signal(Semaphore *sem); Semaphore is a synonym for sem_t , but I find it more readable, and the capital letter reminds me to treat it like an object and pass it by pointer. The implementation of these functions is in sem.c : Semaphore *make_semaphore(int value) { Semaphore *sem = check_malloc(sizeof(Semaphore)); int n = sem_init(sem, 0, value); if (n != 0) perror_exit("sem_init failed"); return sem; } make_semaphore takes the initial value of the semaphore as a parameter. It allocates space for a Semaphore, initializes it, and returns a pointer to Semaphore . sem_init returns 0 if it succeeds and -1 if anything goes wrong. One nice thing about using wrapper functions is that you can encapsulate the error-checking code, which makes the code that uses these functions more readable. Here is the implementation of semaphore_wait : void semaphore_wait(Semaphore *sem) { int n = sem_wait(sem); if (n != 0) perror_exit("sem_wait failed"); } And here is semaphore_signal : void semaphore_signal(Semaphore *sem) { int n = sem_post(sem); if (n != 0) perror_exit("sem_post failed"); } I prefer to call this operation “signal” rather than “post”, although both terms are common.

11.1.1

https://eng.libretexts.org/@go/page/40641

Here’s an example that shows how to use a semaphore as a mutex: Semaphore *mutex = make_semaphore(1); semaphore_wait(mutex); // protected code goes here semaphore_signal(mutex); When you use a semaphore as a mutex, you usually initialize it to 1 to indicate that the mutex is unlocked; that is, one thread can pass the semaphore without blocking. Here I am using the variable name mutex to indicate that the semaphore is being used as a mutex. But remember that the behavior of a semaphore is not the same as a Pthread mutex. This page titled 11.1: POSIX Semaphores is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

11.1.2

https://eng.libretexts.org/@go/page/40641

11.2: Producers and consumers with semaphores Using these semaphore wrapper functions, we can write a solution to the Producer-Consumer problem from Section 10.2. The code in this section is in queue_sem.c . Here’s the new definition of Queue , replacing the mutex and condition variables with semaphores: typedef struct { int *array; int length; int next_in; int next_out; Semaphore *mutex; Semaphore *items; Semaphore *spaces; } Queue;

//-- new //-- new //-- new

And here’s the new version of make_queue : Queue *make_queue(int length) { Queue *queue = (Queue *) malloc(sizeof(Queue)); queue->length = length; queue->array = (int *) malloc(length * sizeof(int)); queue->next_in = 0; queue->next_out = 0; queue->mutex = make_semaphore(1); queue->items = make_semaphore(0); queue->spaces = make_semaphore(length-1); return queue; } mutex is used to guarantee exclusive access to the queue; the initial value is 1, so the mutex is initially unlocked. items is the number of items in the queue, which is also the number of consumer threads that can execute queue_pop without blocking. Initially there are no items in the queue. spaces is the number of empty spaces in the queue, which is the number of producer threads that can execute queue_push without blocking. Initially the number of spaces is the capacity of the queue, which is length-1 , as explained in Section 10.1. Here is the new version of queue_push , which is run by producer threads: void queue_push(Queue *queue, int item) { semaphore_wait(queue->spaces); semaphore_wait(queue->mutex); queue->array[queue->next_in] = item; queue->next_in = queue_incr(queue, queue->next_in); semaphore_signal(queue->mutex);

11.2.1

https://eng.libretexts.org/@go/page/40642

semaphore_signal(queue->items); } Notice that queue_push doesn’t have to call queue_full any more; instead, the semaphore keeps track of how many spaces are available and blocks producers if the queue is full. Here is the new version of queue_pop : int queue_pop(Queue *queue) { semaphore_wait(queue->items); semaphore_wait(queue->mutex); int item = queue->array[queue->next_out]; queue->next_out = queue_incr(queue, queue->next_out); semaphore_signal(queue->mutex); semaphore_signal(queue->spaces); return item; } This solution is explained, using pseudo-code, in Chapter 4 of The Little Book of Semaphores. Using the code in the repository for this book, you should be able to compile and run this solution like this: $ make queue_sem $ ./queue_sem This page titled 11.2: Producers and consumers with semaphores is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

11.2.2

https://eng.libretexts.org/@go/page/40642

11.3: Make your own semaphores Any problem that can be solved with semaphores can also be solved with condition variables and mutexes. We can prove that’s true by using condition variables and mutexes to implement a semaphore. Before you go on, you might want to try this as an exercise: write functions that implement the semaphore API in sem.h using using condition variables and mutexes. In the repository for this book, you’ll find my solution in mysem_soln.c and mysem_soln.h . If you have trouble getting started, you can use the following structure definition, from my solution, as a hint: typedef struct { int value, wakeups; Mutex *mutex; Cond *cond; } Semaphore; value is the value of the semaphore. wakeups counts the number of pending signals; that is, the number of threads that have been woken but have not yet resumed execution. The reason for wakeups is to make sure that our semaphores have Property 3, described in The Little Book of Semaphores. mutex provides exclusive access to value and wakeups ; cond is the condition variable threads wait on if they wait on the semaphore. Here is the initialization code for this structure: Semaphore *make_semaphore(int value) { Semaphore *semaphore = check_malloc(sizeof(Semaphore)); semaphore->value = value; semaphore->wakeups = 0; semaphore->mutex = make_mutex(); semaphore->cond = make_cond(); return semaphore; }

Semaphore implementation Here is my implementation of semaphores using POSIX mutexes and condition variables: void semaphore_wait(Semaphore *semaphore) { mutex_lock(semaphore->mutex); semaphore->value--; if (semaphore->value < 0) { do { cond_wait(semaphore->cond, semaphore->mutex); } while (semaphore->wakeups < 1); semaphore->wakeups--; }

11.3.1

https://eng.libretexts.org/@go/page/40643

mutex_unlock(semaphore->mutex); } When a thread waits on the semaphore, it has to lock the mutex before it decrements value . If the value of the semaphore becomes negative, the thread blocks until a “wakeup” is available. While it is blocked, the mutex is unlocked, so another thread can signal. Here is the code for semaphore_signal : void semaphore_signal(Semaphore *semaphore) { mutex_lock(semaphore->mutex); semaphore->value++; if (semaphore->value wakeups++; cond_signal(semaphore->cond); } mutex_unlock(semaphore->mutex); } Again, a thread has to lock the mutex before it increments value . If the semaphore was negative, that means threads are waiting, so the signalling thread increments wakeups and signals the condition variable. At this point one of the waiting threads might wake up, but the mutex is still locked until the signalling thread unlocks it. At that point, one of the waiting threads returns from cond_wait and checks whether a wakeup is still available. If not, it loops and waits on the condition variable again. If so, it decrements wakeups , unlocks the mutex, and exits. One thing about this solution that might not be obvious is the use of a do...while loop. Can you figure out why it is not a more conventional while loop? What would go wrong? The problem is that with a while loop this implementation would not have Property 3. It would be possible for a thread to signal and then run around and catch its own signal. With the do...while loop, it is guaranteed1 that when a thread signals, one of the waiting threads will get the signal, even if the signaling thread runs around and gets the mutex before one of the waiting threads resumes. 1. Well, almost. It turns out that a well-timed spurious wakeup (see http://en.Wikipedia.org/wiki/Spurious_wakeup) can violate this guarantee. This page titled 11.3: Make your own semaphores is shared under a CC BY-NC license and was authored, remixed, and/or curated by Allen B. Downey (Green Tea Press) .

11.3.2

https://eng.libretexts.org/@go/page/40643

Index A

V

abstraction 2.1: Abstraction and virtualization

Virtualization 2.1: Abstraction and virtualization

1

https://eng.libretexts.org/@go/page/40767

Glossary Sample Word 1 | Sample Definition 1

1

https://eng.libretexts.org/@go/page/43660

Detailed Licensing Overview Title: Think OS - A Brief Introduction to Operating Systems (Downey) Webpages: 80 Applicable Restrictions: Noncommercial All licenses found: CC BY-NC 4.0: 96.3% (77 pages) Undeclared: 3.8% (3 pages)

By Page Think OS - A Brief Introduction to Operating Systems (Downey) - CC BY-NC 4.0

4.5: Everything is a file? - CC BY-NC 4.0 5: More bits and bytes - CC BY-NC 4.0

Front Matter - CC BY-NC 4.0

5.1: Representing integers - CC BY-NC 4.0 5.2: Bitwise operators - CC BY-NC 4.0 5.3: Representing floating-point numbers - CC BY-NC 4.0 5.4: Unions and memory errors - CC BY-NC 4.0 5.5: Representing strings - CC BY-NC 4.0

TitlePage - CC BY-NC 4.0 InfoPage - CC BY-NC 4.0 Table of Contents - Undeclared Licensing - Undeclared 0: Preface - CC BY-NC 4.0

6: Memory management - CC BY-NC 4.0

0.1: Preface - CC BY-NC 4.0 0.2: Using the code - CC BY-NC 4.0 0.3: Contributor List - CC BY-NC 4.0

6.1: Memory errors - CC BY-NC 4.0 6.2: Memory leaks - CC BY-NC 4.0 6.3: Implementation - CC BY-NC 4.0

1: Compilation - CC BY-NC 4.0

7: Caching - CC BY-NC 4.0

1.1: Compiled and interpreted languages - CC BY-NC 4.0 1.2: Static types - CC BY-NC 4.0 1.3: The compilation process - CC BY-NC 4.0 1.4: Object code - CC BY-NC 4.0 1.5: Assembly code - CC BY-NC 4.0 1.6: Preprocessing - CC BY-NC 4.0 1.7: Understanding errors - CC BY-NC 4.0

7.1: How programs run - CC BY-NC 4.0 7.2: Cache performance - CC BY-NC 4.0 7.3: Locality - CC BY-NC 4.0 7.4: Measuring cache performance - CC BY-NC 4.0 7.5: Programming for cache performance - CC BYNC 4.0 7.6: The memory hierarchy - CC BY-NC 4.0 7.7: Caching policy - CC BY-NC 4.0 7.8: Paging - CC BY-NC 4.0

2: Processes - CC BY-NC 4.0 2.1: Abstraction and virtualization - CC BY-NC 4.0 2.2: Isolation - CC BY-NC 4.0 2.3: UNIX processes - CC BY-NC 4.0

8: Multitasking - CC BY-NC 4.0 8.1: Hardware state - CC BY-NC 4.0 8.2: Context switching - CC BY-NC 4.0 8.3: The process life cycle - CC BY-NC 4.0 8.4: Scheduling - CC BY-NC 4.0 8.5: Real-time scheduling - CC BY-NC 4.0

3: Virtual memory - CC BY-NC 4.0 3.1: A bit of information theory - CC BY-NC 4.0 3.2: Memory and storage - CC BY-NC 4.0 3.3: Address spaces - CC BY-NC 4.0 3.4: Memory segments - CC BY-NC 4.0 3.5: Static local variables - CC BY-NC 4.0 3.6: Address translation - CC BY-NC 4.0

9: Threads - CC BY-NC 4.0 9.1: Creating threads - CC BY-NC 4.0 9.2: Creating threads - CC BY-NC 4.0 9.3: Joining threads - CC BY-NC 4.0 9.4: Synchronization errors - CC BY-NC 4.0 9.5: Mutex - CC BY-NC 4.0

4: Files and file systems - CC BY-NC 4.0 4.1: Files and file systems - CC BY-NC 4.0 4.2: Disk performance - CC BY-NC 4.0 4.3: Disk metadata - CC BY-NC 4.0 4.4: Block allocation - CC BY-NC 4.0

10: Condition variables - CC BY-NC 4.0 10.1: The work queue - CC BY-NC 4.0

1

https://eng.libretexts.org/@go/page/96723

10.2: Producers and consumers - CC BY-NC 4.0 10.3: Mutual exclusion - CC BY-NC 4.0 10.4: Condition variables - CC BY-NC 4.0 10.5: Condition variable implementation - CC BY-NC 4.0

11.2: Producers and consumers with semaphores - CC BY-NC 4.0 11.3: Make your own semaphores - CC BY-NC 4.0 Back Matter - CC BY-NC 4.0 Index - CC BY-NC 4.0 Glossary - CC BY-NC 4.0 Detailed Licensing - Undeclared

11: Semaphores in C - CC BY-NC 4.0 11.1: POSIX Semaphores - CC BY-NC 4.0

2

https://eng.libretexts.org/@go/page/96723